Re: [AMBER] job failed for REMD in cluster

From: Jason Swails <>
Date: Mon, 15 Oct 2012 15:47:14 -0400

On Mon, Oct 15, 2012 at 2:53 PM, Albert <> wrote:

> if I have 11, so probably I have to use core like 66, 77?
> which means I cannot run it with CUDA?

You can't run REMD with an odd number of replicas (you will get an error
message to that effect if you try). Run with 10 or 12 instead.

pmemd.MPI (and likewise pmemd.cuda.MPI) does not require the number of
groups to be a multiple of core count, but if you don't do that you need to
explicitly assign each processor to a group. This is an advanced option
and should probably be avoided in general. REMD in Amber is synchronous,
which means if a replica has fewer cores than everyone else and runs
slower, every other replica will wait for the slow one during exchange


Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
AMBER mailing list
Received on Mon Oct 15 2012 - 13:00:08 PDT
Custom Search